Bias in Stats Packages ?

[Dr. David Curtis]

> > 
> > The accuracy of their functions may leave a bit to be desired, though.  
> > I've just starting using Excel 5 and compared a simple linear regression 
> > with the results from Systat.  There was a good bit of difference in the 
> > sums of squares. For instance Excel gave a residual SS of 1501.088 vs 
> > Systat's 1500.565.  The resulting f-ratio was 364.1975 vs 364.300.  I 
> > think I would tend to believe Systat's results more :-)
> > 
> > Dr. Warren L. Kovach              Internet: warrenk at cix.compulink.co.uk
> > 

Sorry I'm picking this up a bit late (I was moving) but am I missing something? Could someone explain to me why these trivial differences are of any practical signficance? I appreciate the point that one might want to know what algorithm a statistical package was claiming to use, but I can't see that the actual results obtained have "a good bit of difference" between them. I mean we're talking statistics here, not physics.


Dave Curtis

Institute of Psychiatry               Janet:       dcurtis at UK.AC.MRC.HGMP
Denmark Hill                          Elsewhere:   dcurtis at HGMP.MRC.AC.UK
London SE5 8AF
Tel 071-603 5411